ゼミナール発表

日時: 09月29日(金)5限(16:50-18:20)


会場: L1

司会: Sakriani Sakti
柴田 大作 1651059: M, 2回目発表 知能コミュニケーション 中村 哲 ☆, 荒牧 英治, 松本 裕治, 田中 宏季

title: Detecting Mild Cognitive Impairment based on Language Abilities

abstract:[Background] In recent years, detecting Mild Cognitive Impairment (MCI) in early stages based on natural language processing (NLP) has drawn much attention. However, there are few studies about the detection of MCI with narratives in Japanese language.[Objective] We propose a method to detect MCI based on linguistic features of Japanese language. [Methods] A total of 39 participants aged 53-90 (mean: 76.89) were divided into two groups based on Mini Mental State Examination (MMSE) scores: The MCI group (n = 10, MMSE = 22-26); The healthy control (HC) group (n = 29, MMSE > 26). [Result] We confirmed that multiple liner regression can classify. [Conclusion] The results suggests that MCI can be detected by using linguistic features.

language of the presentation: Japanese

 
HENTSCHEL MICHAEL ALEXANDER 1661025: D, 中間発表 自然言語処理学 松本 裕治 ☆, 中村 哲, 澤田 宏, 岩田 具治
title: Language Models for Conversational Speech
abstract: In the presentation I am going to talk about modifications to recurrent neural network based language models applicable to rescoring in speech recognition. The first method I am going to present is about the effective use of prosodic information in the context of an imbalance between training text and acoustic training data. The second method will be about cache extensions for neural network based language models. The last modification will be a brief introduction to the topic of domain adaptation, which I am currently working on.
language of the presentation: English
 
城戸 孝士郎 1651042: M, 2回目発表 計算システムズ生物学 金谷 重彦, 中村 哲, MD.ALTAF-UL-AMIN, 小野 直亮

title: The development of a capacitively coupled ECG Monitor system and its use in detecting abnormal electrocardiograms during sleep

abstract: Holter, a medical device for 24-hours electrocardiogram (ECG) measurement, is usually used for abnormal ECG examination such as arrhythmia. However, holter monitoring has the disadvantages such as the necessity of cutaneous electrodes, offline signal process. This is not suitable for long-term monitoring because of skin irritation from direct contact electrodes, and limitation of activity caused by the wired system. Considering the arrhythmia causing sudden death, the series of processes should be performed in real time. Therefore, in this study, we will develop a system that can measure electrocardiogram for a long time and detect arrhythmia in real time by using a capacitively coupled electrocardiograph that does not require direct electrode application.

language of the presentation: Japanese

 

会場: L2

司会: 酒田 信親
佐野 徳美 1651055: M, 2回目発表 生体医用画像 佐藤 嘉伸, 小笠原 司, 大竹 義人, 横田 太
title: Reconstruction of dynamic-3D ultrasound image by integration of static-3D and dynamic-2D for analyzing muscle function

abstract: Analysis of three-dimensional structure of muscles and its movement is important for sports science and rehabilitation. A dynamic-2D ultrasound image is widely used for analyzing muscle function. However, Three-dimensional information outside the imaging plane cannot be observed. On the other hand, static-3D image does not have real-time information by reconstructed ultrasound image.
In this study, we create dynamic-3D volume by integration of static-3D and dynamic-2D for analyzing muscle function. We used slice-to-volume non-rigid registration for dynamic-3D data.In this presentation, we report accuracy of slice-to-volume non-rigid registration by simulation. Furthermore, we discuss result of actual image.

language of the presentation:Japanese

発表題目: 筋機能解析のための静的三次元画像と動的二次元画像の統合による動的三次元画像の再構築

発表概要: 発表概要: スポーツ科学やリハビリテーションの分野において,筋肉内部の構造や動態を解析することは重要である.筋線維動態の解析には一般に,高時間分解能で撮像可能な二次元超音波画像が用いられるが,撮像範囲が平面に限定されるため三次元的な情報は得られない.一方で三次元超音波画像は時間分解能が極めて低く動態を観察することは容易でない. そこで本研究では,筋肉全体の静的三次元画像に対して,座標系の一致した疎な位置での動的二次元画像を,slice-to-volume非剛体位置合わせすることで全体の動的な変形場を予測し動的三次元画像の再構成を行う.また,それによる筋動態解析を目指している.これまでに計測時のスライス位置の再現性を担保するリニアガイドによる撮像装置を発表してきた. 本発表では,次のステップである動態表現における非剛体位置合わせのアルゴリズムの精度をシミュレーション実験により検証するとともに,実画像での初期的検討を行ったので報告する.
 
宮本 康平 1651106: M, 2回目発表 生体医用画像 佐藤 嘉伸, 小笠原 司, 大竹 義人, 横田 太

title: Skeletal muscle fiber tractography using cryosectioned volumes

abstract: Three-dimensional (3D) muscle fiber architecture is important in patient-specific biomechanical simulation. While, it has been analyzed using Diffusion Tensor imaging and ultrasound imaging, these results lack gold standard. Thus, these validations of estimated fiber orientation becomes a problem. We proposed a method for estimating skeletal muscles fiber orientation using high-resolution cryosectioned volumes (Visible Korean Human Dataset) that are observable fine muscle structure. We first calculate structure tensor at each voxel for extract local orientation, and we apply tractography algorithm to structure tensor for obtaining skeletal muscles fiber orientation. In this presentation, we report these result and quantitative evaluation.

language of the presentation: Japanese

 
森口 和也 1651113: M, 2回目発表 生体医用画像 佐藤 嘉伸, 池田 和司, 大竹 義人, 横田 太
title: Estimation of liver fibrois stage using shape feature
abstract: Liver biopsy is used as gold standard diagnostic method of liver fibrosis. However, it is invasive and has risks of sampling error and complications. Therefore, a noninvasive and safe diagnosis method is required. Since the right lobe of liver becomes atrophied and the left lobe hypertrophies with liver fibrosis progress, estimation of fibrosis stage using liver shape can be expected. In this study, we built a linear regression model using shape feature from MR images and report the liber fibrosis stage estimation method with partial least squares regression(PLS-R).
language of the presentation:Japanese